2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
17 #include "pack-revindex.h"
18 #include "sha1-lookup.h"
21 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
22 #define O_NOATIME 01000000
30 static unsigned long sz_fmt(size_t s
) { return (unsigned long)s
; }
33 static size_t sz_fmt(size_t s
) { return s
; }
36 const unsigned char null_sha1
[20];
38 static unsigned int sha1_file_open_flag
= O_NOATIME
;
40 const signed char hexval_table
[256] = {
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
42 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
47 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
48 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
49 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
53 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
62 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
63 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
64 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
65 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
66 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
67 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
68 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
69 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
70 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
71 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
72 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
75 int get_sha1_hex(const char *hex
, unsigned char *sha1
)
78 for (i
= 0; i
< 20; i
++) {
79 unsigned int val
= (hexval(hex
[0]) << 4) | hexval(hex
[1]);
88 static inline int offset_1st_component(const char *path
)
90 if (has_dos_drive_prefix(path
))
91 return 2 + (path
[2] == '/');
95 int safe_create_leading_directories(char *path
)
97 char *pos
= path
+ offset_1st_component(path
);
101 pos
= strchr(pos
, '/');
105 if (!stat(path
, &st
)) {
107 if (!S_ISDIR(st
.st_mode
)) {
112 else if (mkdir(path
, 0777)) {
116 else if (adjust_shared_perm(path
)) {
125 char * sha1_to_hex(const unsigned char *sha1
)
128 static char hexbuffer
[4][50];
129 static const char hex
[] = "0123456789abcdef";
130 char *buffer
= hexbuffer
[3 & ++bufno
], *buf
= buffer
;
133 for (i
= 0; i
< 20; i
++) {
134 unsigned int val
= *sha1
++;
135 *buf
++ = hex
[val
>> 4];
136 *buf
++ = hex
[val
& 0xf];
143 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
146 for (i
= 0; i
< 20; i
++) {
147 static char hex
[] = "0123456789abcdef";
148 unsigned int val
= sha1
[i
];
149 char *pos
= pathbuf
+ i
*2 + (i
> 0);
150 *pos
++ = hex
[val
>> 4];
151 *pos
= hex
[val
& 0xf];
156 * NOTE! This returns a statically allocated buffer, so you have to be
157 * careful about using it. Do an "xstrdup()" if you need to save the
160 * Also note that this returns the location for creating. Reading
161 * SHA1 file can happen from any alternate directory listed in the
162 * DB_ENVIRONMENT environment variable if it is not found in
163 * the primary object database.
165 char *sha1_file_name(const unsigned char *sha1
)
167 static char *name
, *base
;
170 const char *sha1_file_directory
= get_object_directory();
171 int len
= strlen(sha1_file_directory
);
172 base
= xmalloc(len
+ 60);
173 memcpy(base
, sha1_file_directory
, len
);
174 memset(base
+len
, 0, 60);
177 name
= base
+ len
+ 1;
179 fill_sha1_path(name
, sha1
);
183 char *sha1_pack_name(const unsigned char *sha1
)
185 static const char hex
[] = "0123456789abcdef";
186 static char *name
, *base
, *buf
;
190 const char *sha1_file_directory
= get_object_directory();
191 int len
= strlen(sha1_file_directory
);
192 base
= xmalloc(len
+ 60);
193 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory
);
194 name
= base
+ len
+ 11;
199 for (i
= 0; i
< 20; i
++) {
200 unsigned int val
= *sha1
++;
201 *buf
++ = hex
[val
>> 4];
202 *buf
++ = hex
[val
& 0xf];
208 char *sha1_pack_index_name(const unsigned char *sha1
)
210 static const char hex
[] = "0123456789abcdef";
211 static char *name
, *base
, *buf
;
215 const char *sha1_file_directory
= get_object_directory();
216 int len
= strlen(sha1_file_directory
);
217 base
= xmalloc(len
+ 60);
218 sprintf(base
, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory
);
219 name
= base
+ len
+ 11;
224 for (i
= 0; i
< 20; i
++) {
225 unsigned int val
= *sha1
++;
226 *buf
++ = hex
[val
>> 4];
227 *buf
++ = hex
[val
& 0xf];
233 struct alternate_object_database
*alt_odb_list
;
234 static struct alternate_object_database
**alt_odb_tail
;
236 static void read_info_alternates(const char * alternates
, int depth
);
239 * Prepare alternate object database registry.
241 * The variable alt_odb_list points at the list of struct
242 * alternate_object_database. The elements on this list come from
243 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
244 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
245 * whose contents is similar to that environment variable but can be
246 * LF separated. Its base points at a statically allocated buffer that
247 * contains "/the/directory/corresponding/to/.git/objects/...", while
248 * its name points just after the slash at the end of ".git/objects/"
249 * in the example above, and has enough space to hold 40-byte hex
250 * SHA1, an extra slash for the first level indirection, and the
253 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
256 const char *objdir
= get_object_directory();
257 struct alternate_object_database
*ent
;
258 struct alternate_object_database
*alt
;
259 /* 43 = 40-byte + 2 '/' + terminating NUL */
261 int entlen
= pfxlen
+ 43;
264 if (!is_absolute_path(entry
) && relative_base
) {
265 /* Relative alt-odb */
267 base_len
= strlen(relative_base
) + 1;
271 ent
= xmalloc(sizeof(*ent
) + entlen
);
273 if (!is_absolute_path(entry
) && relative_base
) {
274 memcpy(ent
->base
, relative_base
, base_len
- 1);
275 ent
->base
[base_len
- 1] = '/';
276 memcpy(ent
->base
+ base_len
, entry
, len
);
279 memcpy(ent
->base
, entry
, pfxlen
);
281 ent
->name
= ent
->base
+ pfxlen
+ 1;
282 ent
->base
[pfxlen
+ 3] = '/';
283 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
285 /* Detect cases where alternate disappeared */
286 if (stat(ent
->base
, &st
) || !S_ISDIR(st
.st_mode
)) {
287 error("object directory %s does not exist; "
288 "check .git/objects/info/alternates.",
294 /* Prevent the common mistake of listing the same
295 * thing twice, or object directory itself.
297 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
298 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
303 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
308 /* add the alternate entry */
310 alt_odb_tail
= &(ent
->next
);
313 /* recursively add alternates */
314 read_info_alternates(ent
->base
, depth
+ 1);
316 ent
->base
[pfxlen
] = '/';
321 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
322 const char *relative_base
, int depth
)
324 const char *cp
, *last
;
327 error("%s: ignoring alternate object stores, nesting too deep.",
335 if (cp
< ep
&& *cp
== '#') {
336 while (cp
< ep
&& *cp
!= sep
)
341 while (cp
< ep
&& *cp
!= sep
)
344 if (!is_absolute_path(last
) && depth
) {
345 error("%s: ignoring relative alternate object store %s",
346 relative_base
, last
);
348 link_alt_odb_entry(last
, cp
- last
,
349 relative_base
, depth
);
352 while (cp
< ep
&& *cp
== sep
)
358 static void read_info_alternates(const char * relative_base
, int depth
)
363 const char alt_file_name
[] = "info/alternates";
364 /* Given that relative_base is no longer than PATH_MAX,
365 ensure that "path" has enough space to append "/", the
366 file name, "info/alternates", and a trailing NUL. */
367 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
370 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
371 fd
= open(path
, O_RDONLY
);
374 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
378 mapsz
= xsize_t(st
.st_size
);
379 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
382 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
387 void add_to_alternates_file(const char *reference
)
389 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
390 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), 1);
391 char *alt
= mkpath("%s/objects\n", reference
);
392 write_or_die(fd
, alt
, strlen(alt
));
393 if (commit_lock_file(lock
))
394 die("could not close alternates file");
396 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
399 void prepare_alt_odb(void)
406 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
409 alt_odb_tail
= &alt_odb_list
;
410 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
412 read_info_alternates(get_object_directory(), 0);
415 static char *find_sha1_file(const unsigned char *sha1
, struct stat
*st
)
417 char *name
= sha1_file_name(sha1
);
418 struct alternate_object_database
*alt
;
423 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
425 fill_sha1_path(name
, sha1
);
426 if (!stat(alt
->base
, st
))
432 static unsigned int pack_used_ctr
;
433 static unsigned int pack_mmap_calls
;
434 static unsigned int peak_pack_open_windows
;
435 static unsigned int pack_open_windows
;
436 static size_t peak_pack_mapped
;
437 static size_t pack_mapped
;
438 struct packed_git
*packed_git
;
440 void pack_report(void)
443 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
444 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
445 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
446 sz_fmt(getpagesize()),
447 sz_fmt(packed_git_window_size
),
448 sz_fmt(packed_git_limit
));
450 "pack_report: pack_used_ctr = %10u\n"
451 "pack_report: pack_mmap_calls = %10u\n"
452 "pack_report: pack_open_windows = %10u / %10u\n"
453 "pack_report: pack_mapped = "
454 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
457 pack_open_windows
, peak_pack_open_windows
,
458 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
461 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
464 struct pack_idx_header
*hdr
;
466 uint32_t version
, nr
, i
, *index
;
467 int fd
= open(path
, O_RDONLY
);
472 if (fstat(fd
, &st
)) {
476 idx_size
= xsize_t(st
.st_size
);
477 if (idx_size
< 4 * 256 + 20 + 20) {
479 return error("index file %s is too small", path
);
481 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
485 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
486 version
= ntohl(hdr
->idx_version
);
487 if (version
< 2 || version
> 2) {
488 munmap(idx_map
, idx_size
);
489 return error("index file %s is version %d"
490 " and is not supported by this binary"
491 " (try upgrading GIT to a newer version)",
500 index
+= 2; /* skip index header */
501 for (i
= 0; i
< 256; i
++) {
502 uint32_t n
= ntohl(index
[i
]);
504 munmap(idx_map
, idx_size
);
505 return error("non-monotonic index %s", path
);
513 * - 256 index entries 4 bytes each
514 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
515 * - 20-byte SHA1 of the packfile
516 * - 20-byte SHA1 file checksum
518 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
519 munmap(idx_map
, idx_size
);
520 return error("wrong index v1 file size in %s", path
);
522 } else if (version
== 2) {
525 * - 8 bytes of header
526 * - 256 index entries 4 bytes each
527 * - 20-byte sha1 entry * nr
528 * - 4-byte crc entry * nr
529 * - 4-byte offset entry * nr
530 * - 20-byte SHA1 of the packfile
531 * - 20-byte SHA1 file checksum
532 * And after the 4-byte offset table might be a
533 * variable sized table containing 8-byte entries
534 * for offsets larger than 2^31.
536 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
537 unsigned long max_size
= min_size
;
539 max_size
+= (nr
- 1)*8;
540 if (idx_size
< min_size
|| idx_size
> max_size
) {
541 munmap(idx_map
, idx_size
);
542 return error("wrong index v2 file size in %s", path
);
544 if (idx_size
!= min_size
&&
546 * make sure we can deal with large pack offsets.
547 * 31-bit signed offset won't be enough, neither
548 * 32-bit unsigned one will be.
550 (sizeof(off_t
) <= 4)) {
551 munmap(idx_map
, idx_size
);
552 return error("pack too large for current definition of off_t in %s", path
);
556 p
->index_version
= version
;
557 p
->index_data
= idx_map
;
558 p
->index_size
= idx_size
;
563 int open_pack_index(struct packed_git
*p
)
571 idx_name
= xstrdup(p
->pack_name
);
572 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
573 ret
= check_packed_git_idx(idx_name
, p
);
578 static void scan_windows(struct packed_git
*p
,
579 struct packed_git
**lru_p
,
580 struct pack_window
**lru_w
,
581 struct pack_window
**lru_l
)
583 struct pack_window
*w
, *w_l
;
585 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
587 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
597 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
599 struct packed_git
*p
, *lru_p
= NULL
;
600 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
603 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
604 for (p
= packed_git
; p
; p
= p
->next
)
605 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
607 munmap(lru_w
->base
, lru_w
->len
);
608 pack_mapped
-= lru_w
->len
;
610 lru_l
->next
= lru_w
->next
;
612 lru_p
->windows
= lru_w
->next
;
613 if (!lru_p
->windows
&& lru_p
->pack_fd
!= keep_fd
) {
614 close(lru_p
->pack_fd
);
625 void release_pack_memory(size_t need
, int fd
)
627 size_t cur
= pack_mapped
;
628 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
632 void close_pack_windows(struct packed_git
*p
)
635 struct pack_window
*w
= p
->windows
;
638 die("pack '%s' still has open windows to it",
640 munmap(w
->base
, w
->len
);
641 pack_mapped
-= w
->len
;
643 p
->windows
= w
->next
;
648 void unuse_pack(struct pack_window
**w_cursor
)
650 struct pack_window
*w
= *w_cursor
;
658 * Do not call this directly as this leaks p->pack_fd on error return;
659 * call open_packed_git() instead.
661 static int open_packed_git_1(struct packed_git
*p
)
664 struct pack_header hdr
;
665 unsigned char sha1
[20];
666 unsigned char *idx_sha1
;
669 if (!p
->index_data
&& open_pack_index(p
))
670 return error("packfile %s index unavailable", p
->pack_name
);
672 p
->pack_fd
= open(p
->pack_name
, O_RDONLY
);
673 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
676 /* If we created the struct before we had the pack we lack size. */
678 if (!S_ISREG(st
.st_mode
))
679 return error("packfile %s not a regular file", p
->pack_name
);
680 p
->pack_size
= st
.st_size
;
681 } else if (p
->pack_size
!= st
.st_size
)
682 return error("packfile %s size changed", p
->pack_name
);
684 /* We leave these file descriptors open with sliding mmap;
685 * there is no point keeping them open across exec(), though.
687 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
689 return error("cannot determine file descriptor flags");
690 fd_flag
|= FD_CLOEXEC
;
691 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
692 return error("cannot set FD_CLOEXEC");
694 /* Verify we recognize this pack file format. */
695 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
696 return error("file %s is far too short to be a packfile", p
->pack_name
);
697 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
698 return error("file %s is not a GIT packfile", p
->pack_name
);
699 if (!pack_version_ok(hdr
.hdr_version
))
700 return error("packfile %s is version %u and not supported"
701 " (try upgrading GIT to a newer version)",
702 p
->pack_name
, ntohl(hdr
.hdr_version
));
704 /* Verify the pack matches its index. */
705 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
706 return error("packfile %s claims to have %u objects"
707 " while index indicates %u objects",
708 p
->pack_name
, ntohl(hdr
.hdr_entries
),
710 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
711 return error("end of packfile %s is unavailable", p
->pack_name
);
712 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
713 return error("packfile %s signature is unavailable", p
->pack_name
);
714 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
715 if (hashcmp(sha1
, idx_sha1
))
716 return error("packfile %s does not match index", p
->pack_name
);
720 static int open_packed_git(struct packed_git
*p
)
722 if (!open_packed_git_1(p
))
724 if (p
->pack_fd
!= -1) {
731 static int in_window(struct pack_window
*win
, off_t offset
)
733 /* We must promise at least 20 bytes (one hash) after the
734 * offset is available from this window, otherwise the offset
735 * is not actually in this window and a different window (which
736 * has that one hash excess) must be used. This is to support
737 * the object header and delta base parsing routines below.
739 off_t win_off
= win
->offset
;
740 return win_off
<= offset
741 && (offset
+ 20) <= (win_off
+ win
->len
);
744 unsigned char* use_pack(struct packed_git
*p
,
745 struct pack_window
**w_cursor
,
749 struct pack_window
*win
= *w_cursor
;
751 if (p
->pack_fd
== -1 && open_packed_git(p
))
752 die("packfile %s cannot be accessed", p
->pack_name
);
754 /* Since packfiles end in a hash of their content and its
755 * pointless to ask for an offset into the middle of that
756 * hash, and the in_window function above wouldn't match
757 * don't allow an offset too close to the end of the file.
759 if (offset
> (p
->pack_size
- 20))
760 die("offset beyond end of packfile (truncated pack?)");
762 if (!win
|| !in_window(win
, offset
)) {
765 for (win
= p
->windows
; win
; win
= win
->next
) {
766 if (in_window(win
, offset
))
770 size_t window_align
= packed_git_window_size
/ 2;
772 win
= xcalloc(1, sizeof(*win
));
773 win
->offset
= (offset
/ window_align
) * window_align
;
774 len
= p
->pack_size
- win
->offset
;
775 if (len
> packed_git_window_size
)
776 len
= packed_git_window_size
;
777 win
->len
= (size_t)len
;
778 pack_mapped
+= win
->len
;
779 while (packed_git_limit
< pack_mapped
780 && unuse_one_window(p
, p
->pack_fd
))
782 win
->base
= xmmap(NULL
, win
->len
,
783 PROT_READ
, MAP_PRIVATE
,
784 p
->pack_fd
, win
->offset
);
785 if (win
->base
== MAP_FAILED
)
786 die("packfile %s cannot be mapped: %s",
791 if (pack_mapped
> peak_pack_mapped
)
792 peak_pack_mapped
= pack_mapped
;
793 if (pack_open_windows
> peak_pack_open_windows
)
794 peak_pack_open_windows
= pack_open_windows
;
795 win
->next
= p
->windows
;
799 if (win
!= *w_cursor
) {
800 win
->last_used
= pack_used_ctr
++;
804 offset
-= win
->offset
;
806 *left
= win
->len
- xsize_t(offset
);
807 return win
->base
+ offset
;
810 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
813 struct packed_git
*p
= xmalloc(sizeof(*p
) + path_len
+ 2);
816 * Make sure a corresponding .pack file exists and that
817 * the index looks sane.
819 path_len
-= strlen(".idx");
822 memcpy(p
->pack_name
, path
, path_len
);
823 strcpy(p
->pack_name
+ path_len
, ".pack");
824 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
829 /* ok, it looks sane as far as we can check without
830 * actually mapping the pack file.
832 p
->index_version
= 0;
833 p
->index_data
= NULL
;
836 p
->pack_size
= st
.st_size
;
840 p
->pack_local
= local
;
841 p
->mtime
= st
.st_mtime
;
842 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
847 struct packed_git
*parse_pack_index(unsigned char *sha1
)
849 char *path
= sha1_pack_index_name(sha1
);
850 return parse_pack_index_file(sha1
, path
);
853 struct packed_git
*parse_pack_index_file(const unsigned char *sha1
,
854 const char *idx_path
)
856 const char *path
= sha1_pack_name(sha1
);
857 struct packed_git
*p
= xmalloc(sizeof(*p
) + strlen(path
) + 2);
859 if (check_packed_git_idx(idx_path
, p
)) {
864 strcpy(p
->pack_name
, path
);
869 hashcpy(p
->sha1
, sha1
);
873 void install_packed_git(struct packed_git
*pack
)
875 pack
->next
= packed_git
;
879 static void prepare_packed_git_one(char *objdir
, int local
)
881 /* Ensure that this buffer is large enough so that we can
882 append "/pack/" without clobbering the stack even if
883 strlen(objdir) were PATH_MAX. */
884 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
889 sprintf(path
, "%s/pack", objdir
);
894 error("unable to open object pack directory: %s: %s",
895 path
, strerror(errno
));
899 while ((de
= readdir(dir
)) != NULL
) {
900 int namelen
= strlen(de
->d_name
);
901 struct packed_git
*p
;
903 if (!has_extension(de
->d_name
, ".idx"))
906 if (len
+ namelen
+ 1 > sizeof(path
))
909 /* Don't reopen a pack we already have. */
910 strcpy(path
+ len
, de
->d_name
);
911 for (p
= packed_git
; p
; p
= p
->next
) {
912 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
917 /* See if it really is a valid .idx file with corresponding
918 * .pack file that we can map.
920 p
= add_packed_git(path
, len
+ namelen
, local
);
923 install_packed_git(p
);
928 static int sort_pack(const void *a_
, const void *b_
)
930 struct packed_git
*a
= *((struct packed_git
**)a_
);
931 struct packed_git
*b
= *((struct packed_git
**)b_
);
935 * Local packs tend to contain objects specific to our
936 * variant of the project than remote ones. In addition,
937 * remote ones could be on a network mounted filesystem.
938 * Favor local ones for these reasons.
940 st
= a
->pack_local
- b
->pack_local
;
945 * Younger packs tend to contain more recent objects,
946 * and more recent objects tend to get accessed more
949 if (a
->mtime
< b
->mtime
)
951 else if (a
->mtime
== b
->mtime
)
956 static void rearrange_packed_git(void)
958 struct packed_git
**ary
, *p
;
961 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
966 /* prepare an array of packed_git for easier sorting */
967 ary
= xcalloc(n
, sizeof(struct packed_git
*));
968 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
971 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
973 /* link them back again */
974 for (i
= 0; i
< n
- 1; i
++)
975 ary
[i
]->next
= ary
[i
+ 1];
976 ary
[n
- 1]->next
= NULL
;
982 static int prepare_packed_git_run_once
= 0;
983 void prepare_packed_git(void)
985 struct alternate_object_database
*alt
;
987 if (prepare_packed_git_run_once
)
989 prepare_packed_git_one(get_object_directory(), 1);
991 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
993 prepare_packed_git_one(alt
->base
, 0);
996 rearrange_packed_git();
997 prepare_packed_git_run_once
= 1;
1000 void reprepare_packed_git(void)
1002 prepare_packed_git_run_once
= 0;
1003 prepare_packed_git();
1006 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1008 unsigned char real_sha1
[20];
1009 hash_sha1_file(map
, size
, type
, real_sha1
);
1010 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1013 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1018 char *filename
= find_sha1_file(sha1
, &st
);
1024 fd
= open(filename
, O_RDONLY
| sha1_file_open_flag
);
1026 /* See if it works without O_NOATIME */
1027 switch (sha1_file_open_flag
) {
1029 fd
= open(filename
, O_RDONLY
);
1037 /* If it failed once, it will probably fail again.
1038 * Stop using O_NOATIME
1040 sha1_file_open_flag
= 0;
1042 *size
= xsize_t(st
.st_size
);
1043 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1048 static int legacy_loose_object(unsigned char *map
)
1053 * Is it a zlib-compressed buffer? If so, the first byte
1054 * must be 0x78 (15-bit window size, deflated), and the
1055 * first 16-bit word is evenly divisible by 31
1057 word
= (map
[0] << 8) + map
[1];
1058 if (map
[0] == 0x78 && !(word
% 31))
1064 unsigned long unpack_object_header_gently(const unsigned char *buf
, unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1069 unsigned long used
= 0;
1072 *type
= (c
>> 4) & 7;
1078 if (sizeof(long) * 8 <= shift
)
1081 size
+= (c
& 0x7f) << shift
;
1088 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1090 unsigned long size
, used
;
1091 static const char valid_loose_object_type
[8] = {
1093 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1094 0, /* "delta" and others are invalid in a loose object */
1096 enum object_type type
;
1098 /* Get the data stream */
1099 memset(stream
, 0, sizeof(*stream
));
1100 stream
->next_in
= map
;
1101 stream
->avail_in
= mapsize
;
1102 stream
->next_out
= buffer
;
1103 stream
->avail_out
= bufsiz
;
1105 if (legacy_loose_object(map
)) {
1106 inflateInit(stream
);
1107 return inflate(stream
, 0);
1112 * There used to be a second loose object header format which
1113 * was meant to mimic the in-pack format, allowing for direct
1114 * copy of the object data. This format turned up not to be
1115 * really worth it and we don't write it any longer. But we
1116 * can still read it.
1118 used
= unpack_object_header_gently(map
, mapsize
, &type
, &size
);
1119 if (!used
|| !valid_loose_object_type
[type
])
1124 /* Set up the stream for the rest.. */
1125 stream
->next_in
= map
;
1126 stream
->avail_in
= mapsize
;
1127 inflateInit(stream
);
1129 /* And generate the fake traditional header */
1130 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1131 typename(type
), size
);
1135 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1137 int bytes
= strlen(buffer
) + 1;
1138 unsigned char *buf
= xmalloc(1+size
);
1142 n
= stream
->total_out
- bytes
;
1145 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1147 if (bytes
<= size
) {
1149 * The above condition must be (bytes <= size), not
1150 * (bytes < size). In other words, even though we
1151 * expect no more output and set avail_out to zer0,
1152 * the input zlib stream may have bytes that express
1153 * "this concludes the stream", and we *do* want to
1156 * Otherwise we would not be able to test that we
1157 * consumed all the input to reach the expected size;
1158 * we also want to check that zlib tells us that all
1159 * went well with status == Z_STREAM_END at the end.
1161 stream
->next_out
= buf
+ bytes
;
1162 stream
->avail_out
= size
- bytes
;
1163 while (status
== Z_OK
)
1164 status
= inflate(stream
, Z_FINISH
);
1167 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1173 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1174 else if (stream
->avail_in
)
1175 error("garbage at end of loose object '%s'",
1182 * We used to just use "sscanf()", but that's actually way
1183 * too permissive for what we want to check. So do an anal
1184 * object header parse by hand.
1186 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1193 * The type can be at most ten bytes (including the
1194 * terminating '\0' that we add), and is followed by
1203 if (i
>= sizeof(type
))
1209 * The length must follow immediately, and be in canonical
1210 * decimal format (ie "010" is not valid).
1212 size
= *hdr
++ - '0';
1217 unsigned long c
= *hdr
- '0';
1221 size
= size
* 10 + c
;
1227 * The length must be followed by a zero byte
1229 return *hdr
? -1 : type_from_string(type
);
1232 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1238 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1239 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1242 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1245 unsigned long get_size_from_delta(struct packed_git
*p
,
1246 struct pack_window
**w_curs
,
1249 const unsigned char *data
;
1250 unsigned char delta_head
[20], *in
;
1254 memset(&stream
, 0, sizeof(stream
));
1255 stream
.next_out
= delta_head
;
1256 stream
.avail_out
= sizeof(delta_head
);
1258 inflateInit(&stream
);
1260 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1261 stream
.next_in
= in
;
1262 st
= inflate(&stream
, Z_FINISH
);
1263 curpos
+= stream
.next_in
- in
;
1264 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1265 stream
.total_out
< sizeof(delta_head
));
1266 inflateEnd(&stream
);
1267 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
))
1268 die("delta data unpack-initial failed");
1270 /* Examine the initial part of the delta to figure out
1275 /* ignore base size */
1276 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1278 /* Read the result size */
1279 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1282 static off_t
get_delta_base(struct packed_git
*p
,
1283 struct pack_window
**w_curs
,
1285 enum object_type type
,
1286 off_t delta_obj_offset
)
1288 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1291 /* use_pack() assured us we have [base_info, base_info + 20)
1292 * as a range that we can look at without walking off the
1293 * end of the mapped window. Its actually the hash size
1294 * that is assured. An OFS_DELTA longer than the hash size
1295 * is stupid, as then a REF_DELTA would be smaller to store.
1297 if (type
== OBJ_OFS_DELTA
) {
1299 unsigned char c
= base_info
[used
++];
1300 base_offset
= c
& 127;
1303 if (!base_offset
|| MSB(base_offset
, 7))
1304 die("offset value overflow for delta base object");
1305 c
= base_info
[used
++];
1306 base_offset
= (base_offset
<< 7) + (c
& 127);
1308 base_offset
= delta_obj_offset
- base_offset
;
1309 if (base_offset
>= delta_obj_offset
)
1310 die("delta base offset out of bound");
1312 } else if (type
== OBJ_REF_DELTA
) {
1313 /* The base entry _must_ be in the same pack */
1314 base_offset
= find_pack_entry_one(base_info
, p
);
1316 die("failed to find delta-pack base object %s",
1317 sha1_to_hex(base_info
));
1320 die("I am totally screwed");
1324 /* forward declaration for a mutually recursive function */
1325 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1326 unsigned long *sizep
);
1328 static int packed_delta_info(struct packed_git
*p
,
1329 struct pack_window
**w_curs
,
1331 enum object_type type
,
1333 unsigned long *sizep
)
1337 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1338 type
= packed_object_info(p
, base_offset
, NULL
);
1340 /* We choose to only get the type of the base object and
1341 * ignore potentially corrupt pack file that expects the delta
1342 * based on a base with a wrong size. This saves tons of
1346 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1351 static int unpack_object_header(struct packed_git
*p
,
1352 struct pack_window
**w_curs
,
1354 unsigned long *sizep
)
1356 unsigned char *base
;
1359 enum object_type type
;
1361 /* use_pack() assures us we have [base, base + 20) available
1362 * as a range that we can look at at. (Its actually the hash
1363 * size that is assured.) With our object header encoding
1364 * the maximum deflated object size is 2^137, which is just
1365 * insane, so we know won't exceed what we have been given.
1367 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1368 used
= unpack_object_header_gently(base
, left
, &type
, sizep
);
1370 die("object offset outside of pack file");
1376 const char *packed_object_info_detail(struct packed_git
*p
,
1378 unsigned long *size
,
1379 unsigned long *store_size
,
1380 unsigned int *delta_chain_length
,
1381 unsigned char *base_sha1
)
1383 struct pack_window
*w_curs
= NULL
;
1385 unsigned long dummy
;
1386 unsigned char *next_sha1
;
1387 enum object_type type
;
1388 struct revindex_entry
*revidx
;
1390 *delta_chain_length
= 0;
1391 curpos
= obj_offset
;
1392 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1394 revidx
= find_pack_revindex(p
, obj_offset
);
1395 *store_size
= revidx
[1].offset
- obj_offset
;
1400 die("pack %s contains unknown object type %d",
1401 p
->pack_name
, type
);
1406 unuse_pack(&w_curs
);
1407 return typename(type
);
1409 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1410 if (*delta_chain_length
== 0) {
1411 revidx
= find_pack_revindex(p
, obj_offset
);
1412 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1416 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1417 if (*delta_chain_length
== 0)
1418 hashcpy(base_sha1
, next_sha1
);
1419 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1422 (*delta_chain_length
)++;
1423 curpos
= obj_offset
;
1424 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1428 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1429 unsigned long *sizep
)
1431 struct pack_window
*w_curs
= NULL
;
1433 off_t curpos
= obj_offset
;
1434 enum object_type type
;
1436 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1441 type
= packed_delta_info(p
, &w_curs
, curpos
,
1442 type
, obj_offset
, sizep
);
1452 die("pack %s contains unknown object type %d",
1453 p
->pack_name
, type
);
1455 unuse_pack(&w_curs
);
1459 static void *unpack_compressed_entry(struct packed_git
*p
,
1460 struct pack_window
**w_curs
,
1466 unsigned char *buffer
, *in
;
1468 buffer
= xmalloc(size
+ 1);
1470 memset(&stream
, 0, sizeof(stream
));
1471 stream
.next_out
= buffer
;
1472 stream
.avail_out
= size
;
1474 inflateInit(&stream
);
1476 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1477 stream
.next_in
= in
;
1478 st
= inflate(&stream
, Z_FINISH
);
1479 curpos
+= stream
.next_in
- in
;
1480 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1481 inflateEnd(&stream
);
1482 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1490 #define MAX_DELTA_CACHE (256)
1492 static size_t delta_base_cached
;
1494 static struct delta_base_cache_lru_list
{
1495 struct delta_base_cache_lru_list
*prev
;
1496 struct delta_base_cache_lru_list
*next
;
1497 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1499 static struct delta_base_cache_entry
{
1500 struct delta_base_cache_lru_list lru
;
1502 struct packed_git
*p
;
1505 enum object_type type
;
1506 } delta_base_cache
[MAX_DELTA_CACHE
];
1508 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1512 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1513 hash
+= (hash
>> 8) + (hash
>> 16);
1514 return hash
% MAX_DELTA_CACHE
;
1517 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1518 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1521 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1522 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1525 if (ret
&& ent
->p
== p
&& ent
->base_offset
== base_offset
)
1526 goto found_cache_entry
;
1527 return unpack_entry(p
, base_offset
, type
, base_size
);
1532 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1533 ent
->lru
.prev
->next
= ent
->lru
.next
;
1534 delta_base_cached
-= ent
->size
;
1536 ret
= xmemdupz(ent
->data
, ent
->size
);
1539 *base_size
= ent
->size
;
1543 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1548 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1549 ent
->lru
.prev
->next
= ent
->lru
.next
;
1550 delta_base_cached
-= ent
->size
;
1554 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1555 void *base
, unsigned long base_size
, enum object_type type
)
1557 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1558 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1559 struct delta_base_cache_lru_list
*lru
;
1561 release_delta_base_cache(ent
);
1562 delta_base_cached
+= base_size
;
1564 for (lru
= delta_base_cache_lru
.next
;
1565 delta_base_cached
> delta_base_cache_limit
1566 && lru
!= &delta_base_cache_lru
;
1568 struct delta_base_cache_entry
*f
= (void *)lru
;
1569 if (f
->type
== OBJ_BLOB
)
1570 release_delta_base_cache(f
);
1572 for (lru
= delta_base_cache_lru
.next
;
1573 delta_base_cached
> delta_base_cache_limit
1574 && lru
!= &delta_base_cache_lru
;
1576 struct delta_base_cache_entry
*f
= (void *)lru
;
1577 release_delta_base_cache(f
);
1581 ent
->base_offset
= base_offset
;
1584 ent
->size
= base_size
;
1585 ent
->lru
.next
= &delta_base_cache_lru
;
1586 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1587 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1588 delta_base_cache_lru
.prev
= &ent
->lru
;
1591 static void *unpack_delta_entry(struct packed_git
*p
,
1592 struct pack_window
**w_curs
,
1594 unsigned long delta_size
,
1596 enum object_type
*type
,
1597 unsigned long *sizep
)
1599 void *delta_data
, *result
, *base
;
1600 unsigned long base_size
;
1603 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1604 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1606 die("failed to read delta base object"
1607 " at %"PRIuMAX
" from %s",
1608 (uintmax_t)base_offset
, p
->pack_name
);
1610 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1612 die("failed to unpack compressed delta"
1613 " at %"PRIuMAX
" from %s",
1614 (uintmax_t)curpos
, p
->pack_name
);
1615 result
= patch_delta(base
, base_size
,
1616 delta_data
, delta_size
,
1619 die("failed to apply delta");
1621 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1625 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1626 enum object_type
*type
, unsigned long *sizep
)
1628 struct pack_window
*w_curs
= NULL
;
1629 off_t curpos
= obj_offset
;
1632 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1636 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1637 obj_offset
, type
, sizep
);
1643 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1646 die("unknown object type %i in %s", *type
, p
->pack_name
);
1648 unuse_pack(&w_curs
);
1652 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1655 const unsigned char *index
= p
->index_data
;
1657 if (open_pack_index(p
))
1659 index
= p
->index_data
;
1661 if (n
>= p
->num_objects
)
1664 if (p
->index_version
== 1) {
1665 return index
+ 24 * n
+ 4;
1668 return index
+ 20 * n
;
1672 static off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1674 const unsigned char *index
= p
->index_data
;
1676 if (p
->index_version
== 1) {
1677 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1680 index
+= 8 + p
->num_objects
* (20 + 4);
1681 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1682 if (!(off
& 0x80000000))
1684 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1685 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1686 ntohl(*((uint32_t *)(index
+ 4)));
1690 off_t
find_pack_entry_one(const unsigned char *sha1
,
1691 struct packed_git
*p
)
1693 const uint32_t *level1_ofs
= p
->index_data
;
1694 const unsigned char *index
= p
->index_data
;
1695 unsigned hi
, lo
, stride
;
1696 static int use_lookup
= -1;
1697 static int debug_lookup
= -1;
1699 if (debug_lookup
< 0)
1700 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1703 if (open_pack_index(p
))
1705 level1_ofs
= p
->index_data
;
1706 index
= p
->index_data
;
1708 if (p
->index_version
> 1) {
1713 hi
= ntohl(level1_ofs
[*sha1
]);
1714 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
1715 if (p
->index_version
> 1) {
1723 printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1724 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
1727 use_lookup
= !!getenv("GIT_USE_LOOKUP");
1729 int pos
= sha1_entry_pos(index
, stride
, 0,
1730 lo
, hi
, p
->num_objects
, sha1
);
1733 return nth_packed_object_offset(p
, pos
);
1737 unsigned mi
= (lo
+ hi
) / 2;
1738 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
1741 printf("lo %u hi %u rg %u mi %u\n",
1742 lo
, hi
, hi
- lo
, mi
);
1744 return nth_packed_object_offset(p
, mi
);
1753 int matches_pack_name(struct packed_git
*p
, const char *name
)
1755 const char *last_c
, *c
;
1757 if (!strcmp(p
->pack_name
, name
))
1760 for (c
= p
->pack_name
, last_c
= c
; *c
;)
1765 if (!strcmp(last_c
, name
))
1771 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
, const char **ignore_packed
)
1773 static struct packed_git
*last_found
= (void *)1;
1774 struct packed_git
*p
;
1777 prepare_packed_git();
1780 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
1783 if (ignore_packed
) {
1785 for (ig
= ignore_packed
; *ig
; ig
++)
1786 if (matches_pack_name(p
, *ig
))
1792 offset
= find_pack_entry_one(sha1
, p
);
1795 * We are about to tell the caller where they can
1796 * locate the requested object. We better make
1797 * sure the packfile is still here and can be
1798 * accessed before supplying that answer, as
1799 * it may have been deleted since the index
1802 if (p
->pack_fd
== -1 && open_packed_git(p
)) {
1803 error("packfile %s cannot be accessed", p
->pack_name
);
1808 hashcpy(e
->sha1
, sha1
);
1814 if (p
== last_found
)
1818 if (p
== last_found
)
1824 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
1825 struct packed_git
*packs
)
1827 struct packed_git
*p
;
1829 for (p
= packs
; p
; p
= p
->next
) {
1830 if (find_pack_entry_one(sha1
, p
))
1837 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1840 unsigned long mapsize
, size
;
1845 map
= map_sha1_file(sha1
, &mapsize
);
1847 return error("unable to find %s", sha1_to_hex(sha1
));
1848 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
1849 status
= error("unable to unpack %s header",
1851 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
1852 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
1855 inflateEnd(&stream
);
1856 munmap(map
, mapsize
);
1860 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
1862 struct pack_entry e
;
1864 if (!find_pack_entry(sha1
, &e
, NULL
)) {
1865 reprepare_packed_git();
1866 if (!find_pack_entry(sha1
, &e
, NULL
))
1867 return sha1_loose_object_info(sha1
, sizep
);
1869 return packed_object_info(e
.p
, e
.offset
, sizep
);
1872 static void *read_packed_sha1(const unsigned char *sha1
,
1873 enum object_type
*type
, unsigned long *size
)
1875 struct pack_entry e
;
1877 if (!find_pack_entry(sha1
, &e
, NULL
))
1880 return cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
1884 * This is meant to hold a *small* number of objects that you would
1885 * want read_sha1_file() to be able to return, but yet you do not want
1886 * to write them into the object store (e.g. a browse-only
1889 static struct cached_object
{
1890 unsigned char sha1
[20];
1891 enum object_type type
;
1895 static int cached_object_nr
, cached_object_alloc
;
1897 static struct cached_object empty_tree
= {
1898 /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1899 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1900 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1906 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
1909 struct cached_object
*co
= cached_objects
;
1911 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
1912 if (!hashcmp(co
->sha1
, sha1
))
1915 if (!hashcmp(sha1
, empty_tree
.sha1
))
1920 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
1921 unsigned char *sha1
)
1923 struct cached_object
*co
;
1925 hash_sha1_file(buf
, len
, typename(type
), sha1
);
1926 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
1928 if (cached_object_alloc
<= cached_object_nr
) {
1929 cached_object_alloc
= alloc_nr(cached_object_alloc
);
1930 cached_objects
= xrealloc(cached_objects
,
1931 sizeof(*cached_objects
) *
1932 cached_object_alloc
);
1934 co
= &cached_objects
[cached_object_nr
++];
1937 co
->buf
= xmalloc(len
);
1938 memcpy(co
->buf
, buf
, len
);
1939 hashcpy(co
->sha1
, sha1
);
1943 void *read_sha1_file(const unsigned char *sha1
, enum object_type
*type
,
1944 unsigned long *size
)
1946 unsigned long mapsize
;
1948 struct cached_object
*co
;
1950 co
= find_cached_object(sha1
);
1954 return xmemdupz(co
->buf
, co
->size
);
1957 buf
= read_packed_sha1(sha1
, type
, size
);
1960 map
= map_sha1_file(sha1
, &mapsize
);
1962 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
1963 munmap(map
, mapsize
);
1966 reprepare_packed_git();
1967 return read_packed_sha1(sha1
, type
, size
);
1970 void *read_object_with_reference(const unsigned char *sha1
,
1971 const char *required_type_name
,
1972 unsigned long *size
,
1973 unsigned char *actual_sha1_return
)
1975 enum object_type type
, required_type
;
1977 unsigned long isize
;
1978 unsigned char actual_sha1
[20];
1980 required_type
= type_from_string(required_type_name
);
1981 hashcpy(actual_sha1
, sha1
);
1983 int ref_length
= -1;
1984 const char *ref_type
= NULL
;
1986 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
1989 if (type
== required_type
) {
1991 if (actual_sha1_return
)
1992 hashcpy(actual_sha1_return
, actual_sha1
);
1995 /* Handle references */
1996 else if (type
== OBJ_COMMIT
)
1998 else if (type
== OBJ_TAG
)
1999 ref_type
= "object ";
2004 ref_length
= strlen(ref_type
);
2006 if (ref_length
+ 40 > isize
||
2007 memcmp(buffer
, ref_type
, ref_length
) ||
2008 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2013 /* Now we have the ID of the referred-to object in
2014 * actual_sha1. Check again. */
2018 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2019 const char *type
, unsigned char *sha1
,
2020 char *hdr
, int *hdrlen
)
2024 /* Generate the header */
2025 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2029 SHA1_Update(&c
, hdr
, *hdrlen
);
2030 SHA1_Update(&c
, buf
, len
);
2031 SHA1_Final(sha1
, &c
);
2035 * Link the tempfile to the final place, possibly creating the
2036 * last directory level as you do so.
2038 * Returns the errno on failure, 0 on success.
2040 static int link_temp_to_file(const char *tmpfile
, const char *filename
)
2045 if (!link(tmpfile
, filename
))
2049 * Try to mkdir the last path component if that failed.
2051 * Re-try the "link()" regardless of whether the mkdir
2052 * succeeds, since a race might mean that somebody
2056 dir
= strrchr(filename
, '/');
2059 if (!mkdir(filename
, 0777) && adjust_shared_perm(filename
)) {
2064 if (!link(tmpfile
, filename
))
2072 * Move the just written object into its final resting place
2074 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2076 int ret
= link_temp_to_file(tmpfile
, filename
);
2079 * Coda hack - coda doesn't like cross-directory links,
2080 * so we fall back to a rename, which will mean that it
2081 * won't be able to check collisions, but that's not a
2084 * The same holds for FAT formatted media.
2086 * When this succeeds, we just return 0. We have nothing
2089 if (ret
&& ret
!= EEXIST
) {
2090 if (!rename(tmpfile
, filename
))
2096 if (ret
!= EEXIST
) {
2097 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2099 /* FIXME!!! Collision check here ? */
2105 static int write_buffer(int fd
, const void *buf
, size_t len
)
2107 if (write_in_full(fd
, buf
, len
) < 0)
2108 return error("file write error (%s)", strerror(errno
));
2112 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2113 unsigned char *sha1
)
2117 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2121 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2122 void *buf
, unsigned long len
, time_t mtime
)
2125 unsigned char *compressed
;
2128 static char tmpfile
[PATH_MAX
];
2130 filename
= sha1_file_name(sha1
);
2131 fd
= open(filename
, O_RDONLY
);
2134 * FIXME!!! We might do collision checking here, but we'd
2135 * need to uncompress the old file and check it. Later.
2141 if (errno
!= ENOENT
) {
2142 return error("sha1 file %s: %s\n", filename
, strerror(errno
));
2145 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2147 fd
= mkstemp(tmpfile
);
2150 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2152 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2156 memset(&stream
, 0, sizeof(stream
));
2157 deflateInit(&stream
, zlib_compression_level
);
2158 size
= 8 + deflateBound(&stream
, len
+hdrlen
);
2159 compressed
= xmalloc(size
);
2162 stream
.next_out
= compressed
;
2163 stream
.avail_out
= size
;
2165 /* First header.. */
2166 stream
.next_in
= (unsigned char *)hdr
;
2167 stream
.avail_in
= hdrlen
;
2168 while (deflate(&stream
, 0) == Z_OK
)
2171 /* Then the data itself.. */
2172 stream
.next_in
= buf
;
2173 stream
.avail_in
= len
;
2174 ret
= deflate(&stream
, Z_FINISH
);
2175 if (ret
!= Z_STREAM_END
)
2176 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2178 ret
= deflateEnd(&stream
);
2180 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2182 size
= stream
.total_out
;
2184 if (write_buffer(fd
, compressed
, size
) < 0)
2185 die("unable to write sha1 file");
2188 die("unable to write sha1 file");
2194 utb
.modtime
= mtime
;
2195 if (utime(tmpfile
, &utb
) < 0)
2196 warning("failed utime() on %s: %s",
2197 tmpfile
, strerror(errno
));
2200 return move_temp_to_file(tmpfile
, filename
);
2203 int write_sha1_file(void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2205 unsigned char sha1
[20];
2209 /* Normally if we have it in the pack then we do not bother writing
2210 * it out into .git/objects/??/?{38} file.
2212 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2214 hashcpy(returnsha1
, sha1
);
2215 if (has_sha1_file(sha1
))
2217 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2220 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2225 enum object_type type
;
2229 if (find_sha1_file(sha1
, &st
))
2231 buf
= read_packed_sha1(sha1
, &type
, &len
);
2233 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2234 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2235 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2239 * We need to unpack and recompress the object for writing
2240 * it out to a different file.
2242 static void *repack_object(const unsigned char *sha1
, unsigned long *objsize
)
2246 unsigned char *unpacked
;
2248 enum object_type type
;
2253 /* need to unpack and recompress it by itself */
2254 unpacked
= read_packed_sha1(sha1
, &type
, &len
);
2256 error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2258 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2261 memset(&stream
, 0, sizeof(stream
));
2262 deflateInit(&stream
, zlib_compression_level
);
2263 size
= deflateBound(&stream
, len
+ hdrlen
);
2264 buf
= xmalloc(size
);
2267 stream
.next_out
= buf
;
2268 stream
.avail_out
= size
;
2270 /* First header.. */
2271 stream
.next_in
= (void *)hdr
;
2272 stream
.avail_in
= hdrlen
;
2273 while (deflate(&stream
, 0) == Z_OK
)
2276 /* Then the data itself.. */
2277 stream
.next_in
= unpacked
;
2278 stream
.avail_in
= len
;
2279 while (deflate(&stream
, Z_FINISH
) == Z_OK
)
2281 deflateEnd(&stream
);
2284 *objsize
= stream
.total_out
;
2288 int write_sha1_to_fd(int fd
, const unsigned char *sha1
)
2291 unsigned long objsize
;
2292 void *buf
= map_sha1_file(sha1
, &objsize
);
2295 retval
= write_buffer(fd
, buf
, objsize
);
2296 munmap(buf
, objsize
);
2300 buf
= repack_object(sha1
, &objsize
);
2301 retval
= write_buffer(fd
, buf
, objsize
);
2306 int write_sha1_from_fd(const unsigned char *sha1
, int fd
, char *buffer
,
2307 size_t bufsize
, size_t *bufposn
)
2309 char tmpfile
[PATH_MAX
];
2312 unsigned char real_sha1
[20];
2313 unsigned char discard
[4096];
2317 snprintf(tmpfile
, sizeof(tmpfile
), "%s/tmp_obj_XXXXXX", get_object_directory());
2319 local
= mkstemp(tmpfile
);
2322 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2324 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2327 memset(&stream
, 0, sizeof(stream
));
2329 inflateInit(&stream
);
2336 stream
.avail_in
= *bufposn
;
2337 stream
.next_in
= (unsigned char *) buffer
;
2339 stream
.next_out
= discard
;
2340 stream
.avail_out
= sizeof(discard
);
2341 ret
= inflate(&stream
, Z_SYNC_FLUSH
);
2342 SHA1_Update(&c
, discard
, sizeof(discard
) -
2344 } while (stream
.avail_in
&& ret
== Z_OK
);
2345 if (write_buffer(local
, buffer
, *bufposn
- stream
.avail_in
) < 0)
2346 die("unable to write sha1 file");
2347 memmove(buffer
, buffer
+ *bufposn
- stream
.avail_in
,
2349 *bufposn
= stream
.avail_in
;
2353 size
= xread(fd
, buffer
+ *bufposn
, bufsize
- *bufposn
);
2358 return error("Connection closed?");
2359 perror("Reading from connection");
2364 inflateEnd(&stream
);
2366 fchmod(local
, 0444);
2367 if (close(local
) != 0)
2368 die("unable to write sha1 file");
2369 SHA1_Final(real_sha1
, &c
);
2370 if (ret
!= Z_STREAM_END
) {
2372 return error("File %s corrupted", sha1_to_hex(sha1
));
2374 if (hashcmp(sha1
, real_sha1
)) {
2376 return error("File %s has bad hash", sha1_to_hex(sha1
));
2379 return move_temp_to_file(tmpfile
, sha1_file_name(sha1
));
2382 int has_pack_index(const unsigned char *sha1
)
2385 if (stat(sha1_pack_index_name(sha1
), &st
))
2390 int has_pack_file(const unsigned char *sha1
)
2393 if (stat(sha1_pack_name(sha1
), &st
))
2398 int has_sha1_pack(const unsigned char *sha1
, const char **ignore_packed
)
2400 struct pack_entry e
;
2401 return find_pack_entry(sha1
, &e
, ignore_packed
);
2404 int has_sha1_file(const unsigned char *sha1
)
2407 struct pack_entry e
;
2409 if (find_pack_entry(sha1
, &e
, NULL
))
2411 return find_sha1_file(sha1
, &st
) ? 1 : 0;
2414 int index_pipe(unsigned char *sha1
, int fd
, const char *type
, int write_object
)
2419 strbuf_init(&buf
, 0);
2420 if (strbuf_read(&buf
, fd
, 4096) < 0) {
2421 strbuf_release(&buf
);
2428 ret
= write_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2430 ret
= hash_sha1_file(buf
.buf
, buf
.len
, type
, sha1
);
2431 strbuf_release(&buf
);
2436 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2437 enum object_type type
, const char *path
)
2439 size_t size
= xsize_t(st
->st_size
);
2441 int ret
, re_allocated
= 0;
2444 buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2451 * Convert blobs to git internal format
2453 if ((type
== OBJ_BLOB
) && S_ISREG(st
->st_mode
)) {
2455 strbuf_init(&nbuf
, 0);
2456 if (convert_to_git(path
, buf
, size
, &nbuf
,
2457 write_object
? safe_crlf
: 0)) {
2459 buf
= strbuf_detach(&nbuf
, &size
);
2465 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2467 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2477 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2483 switch (st
->st_mode
& S_IFMT
) {
2485 fd
= open(path
, O_RDONLY
);
2487 return error("open(\"%s\"): %s", path
,
2489 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
) < 0)
2490 return error("%s: failed to insert into database",
2494 len
= xsize_t(st
->st_size
);
2495 target
= xmalloc(len
+ 1);
2496 if (readlink(path
, target
, len
+ 1) != st
->st_size
) {
2497 char *errstr
= strerror(errno
);
2499 return error("readlink(\"%s\"): %s", path
,
2503 hash_sha1_file(target
, len
, blob_type
, sha1
);
2504 else if (write_sha1_file(target
, len
, blob_type
, sha1
))
2505 return error("%s: failed to insert into database",
2510 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2512 return error("%s: unsupported file type", path
);
2517 int read_pack_header(int fd
, struct pack_header
*header
)
2519 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2520 /* "eof before pack header was fully read" */
2521 return PH_ERROR_EOF
;
2523 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2524 /* "protocol error (pack signature mismatch detected)" */
2525 return PH_ERROR_PACK_SIGNATURE
;
2526 if (!pack_version_ok(header
->hdr_version
))
2527 /* "protocol error (pack version unsupported)" */
2528 return PH_ERROR_PROTOCOL
;